PolyBench - Property Comparison & Ranking - Ranking Kendall Tau: leaderboard
Ranking questions from PolyBench (Mohanty et al., Stony Brook University, 2026) that ask the model to order held-out polymers by a property value. The score is Kendall's tau between the predicted and the reference order, from -1 to 1; higher is better. Models suffixed PolyBench are the authors' fine-tunes on the benchmark's training split.
Metric: Kendall's Tau (-1 to 1; predicted vs reference property ranking). Source: arxiv.org. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 0.87 | #121 |
| 2 | GPT-4o | 0.86 | #741 |
| 3 | Claude Sonnet 4.6 | 0.85 | #158 |
| 4 | Phi-4 | 0.7 | #1920 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=polybench-property-comparison-ranking-ranking-kendall-tau · How It Works · Data refreshed daily, snapshot 2026-11-12.